[singhrac]

Sure, but if Nvidia/OpenAI/Google/Facebook have shown anything, it's that there's always more kernels to invent and train bigger nets with.